Research map / Image generation / GANs
No Modes Left Behind: Capturing the Data Distribution Effectively Using GANs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace the vanilla BIGAN objective with LOGAN lol2 | BIGAN · CIFAR10 | Inception score: 5.31 ± .07 -> 6.23 ± .03 | Helped |
| replace the vanilla BIGAN objective with LOGAN lol1 | BIGAN · CIFAR10 | Inception score: 5.31 ± .07 -> 6.11 ± .05 | Helped |
| replace the Wasserstein objective with LOGAN lol2 | BIGAN · CIFAR10 | Inception score: 4.8 ± .15 -> 6.23 ± .03 | Helped |
| replace the vanilla BIGAN objective with LOGAN lol1 | BIGAN · BIGAN reconstruction | SSIM: 0.223 ± .08 -> 0.238 ± .09 | Helped |
| replace the vanilla BIGAN objective with LOGAN lol2 | BIGAN · BIGAN reconstruction | SSIM: 0.223 ± .08 -> 0.230 ± .09 | No clear effect |
| use lol2 squared-error discriminator loss instead of lol1 log-likelihood loss (ablation) | LOGAN BIGAN · CIFAR10 | Inception score: 6.11 ± .05 -> 6.23 ± .03 | Helped |
| use lol2 squared-error discriminator loss instead of lol1 log-likelihood loss (ablation) | LOGAN DCGAN · CIFAR10 | Inception score: 6.52 ± .15 -> 6.55 ± .19 | No clear effect |
| add pair-wise gradient penalty (ablation) | LOGAN BIGAN · 2D toy datasets | mode coverage | Helped |
| add pair-wise gradient penalty (ablation) | BIGAN · 2D toy datasets | mode coverage | Helped |
| replace Wasserstein estimate with LOGAN logit loss | BIGAN · 2D toy datasets | mode coverage | Mixed |
| remove batch normalization from the discriminator (ablation) | all loss cases · image datasets | training performance | Helped |
| add direct reconstruction mean-squared-error modeling objectives (ablation) | BIGAN encoder and generator · mode capture | mode coverage | Hurt |
Closest papers by meaning
- Mode Regularized Generative Adversarial Networks (ICLR 2017 · similarity 0.91)
- Training Generative Adversarial Networks from Incomplete Observations using Factorised Discriminators (ICLR 2020 · similarity 0.89)
- Structure-preserving GANs (ICML 2022 · similarity 0.89)
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · similarity 0.89)
- MGAN: Training Generative Adversarial Nets with Multiple Generators (ICLR 2018 · similarity 0.89)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.